Chapter 10 Introduction Basic cog function is to categorize We have names for groups of things such as dogs, cats,...

..etc We can combine single words to create an unlimited number of new categories Without categories we would be unable to make any sense of our exp or to profit from it We would be lost in a sea of new exp helpless to employ any of our prior knowledge to navigate it If we conceptualize every little thing we would have no basis whatsoever for making predictions and selecting treatments Similarity is a very important principle of categorization Categorizing applies wherever relevant knowledge might be brought to bear Computational Complexity o Easy to show that computational quickly runs into problems of comp complex o We necessarily employ a minuscule subset of these possibilities o We can determine that the number of ways assigning those things to categories increases very rapidly with n o Why we have categories , one answer is that our categories mirror the structure of the world. o Perhaps the world comes organized into natural clusters o Our concepts mirror those natural groupings or categories Functions of Concepts o We use concept to refer to a mental rep and category to refer to the set of entities or examples picked out by the concept o Suggestion that categories have an independent existence in the world o This previous view is WRONG because we view categories as something that we create by the human mind as it relates to the world o Classification - Concepts allow us to treat discriminable different things as equivalent Classification = deciding that two or more things are members of the same category o Understanding - Classification allows intelligent orgs to break apart their exp in to meaningful chunks and to construct an interpretation of it o Concepts also support explanations ; understanding why a friend reacted to a stick with alarm is explained with the knowledge that he or she initially classified it as a rattlesnake o Prediction - Allows one to make predictions concerning the future, predications that can be used to select plans and actions o Reasoning - Concepts support reasoning - You don't need to store every fact or probability if inferences can be derived from info that is stored People can combine concepts to describe novel situations and to envision future states of affairs o Communication - People share knowledge and index it in terms of the same categories, they will be able to communicate with each other Communication allows learning on the basis of indirect experience o One can readily see that concepts function in multiple ways and are essential to mental life Concepts and Misconceptions o Our categories are a sign that some set of items should be treated equivalently in some way o We may end up treating members of a category as similar based on their category membership o There are labels made that make examples within a category more similar and the difference across categories more distinctive

Chapter 10
o o o

When people are brought into labs and put into groups based on meaningless dimension, ingroup favoritism is the result Despite grouping being arbitrary, people gave out-groups more of the negative behaviors than in-group Thinking about race is organized by theories and beliefs about innate potential

Structure Of Natural Object Concepts Similar things tend to belong to the same category and dissimilar things tend to belong to different categories Classical View - Assumes that concepts have defining features that act like criteria or rules for determining category membership o All members of this category are equally good examples of it and learning involves these definable features o People fail to come up with defining features when asked and they do not necessarily agree with each other o Most current natural concepts do not fit the classical view o Although some concepts like triangle may be well defined, many concepts do not appear to be, and for this reason cognitive psychologists have pretty much abandoned the classic view Probabilistic View - "Prototype View" - Argues that concepts are organized around properties that are characteristic or typical of category members but crucially, they need not be true of all members o The features are probable o If categories are organized around characteristic properties than some may have more properties than others o Category boundaries may be fuzzy o Learning about a category cannot be equated with determining what the defining eaters are because there may not be any o Features and Typicality Listed properties for some members occurred frequently in other members whereas other members had properties that occurred less frequently So to correctly categorize beats and penguins we would have to assign extra weight to the feature "has feathers" for birds More frequently a category member's properties appeared within a category the higher was its rated typicality for that category Internal structure of goal-derived categories such as "things to wear in the snow" is determined by some ideal associated with the category Ideals will only come into play when the category of interest lacks the natural similarity structure that characterizes common taxonomic categories such as bird Internal structure of category tree as organized around the positive ideal of height Differences in goodness of example effects appear to depend on expertise, not the wording of the instructions People who have considerable knowledge in a domain tend to base typicality judgments on ideals not the number of typical features The internal structure of categories is determined not only by classification processes but also by other conceptual functions o Mental Reps of Fuzzy Categories Family Resemblance - Tend to share features such as a distinctive chin Prototype = Example that possessed all the characteristics features of a category

Chapter 10 Classifying a new example is done by comparing the new item to the prototype If candidate is example is similar enough to the prototype for a category it is classified as member of category Based on exp with examples of the category, people learn the central tendency or prototype that becomes the summary mental rep for the category A prototype is like a stereotype which also may be true of no individual Many variables relevant to human category learning are correlated with each other, and it is hard opt determine which variables really are important One can isolate some variable or set of variables of interest by breaking these natural correlations People are learning these fuzzy categories by learning or forming prototypes Detailed analysis show problems with prototype sass mental reps A prototype rep discards info concerning category size , variability of examples, and correlations among tributes A prototype does not capture this awareness of correlational info How well an item belongs to a concept depends on the context in which it is presented People seem to adjust their expectations in a manner sensitive to different settings Prototype reps seem to discard to much info that can be shown to be relevant to human categorization Every model for category learning has some constraints or biases associated with it in the sense of predicting that some kind sofa classification problems should be easier to master than others Linear Reparability = Categories are linearly separable if one can categorize the ex perfectly by adding and weighting the evidence from individual features Prototype models imply that categories must be linearly separable to be learnable Summing of evidence against a criterion when classifying ex on the basis of similarity Categories are separable if there is a straight line ^ that perfectly separates the categories, if no line will separate categories then there will be no way to construct prototypes People should find it easier to learn categories that are linearly separable than categories that are not linearly separable For social categories there is strong evidence that linear severability does matter People may have a set of schemas or thematic structure that facilitate the integration of info across dimensions People show considerable flex in relating features of ex to more abstract underlying properties This linear constraint implied by prototype models doesn't seem to hold too well Exemplar Theories Provide an alternative way of rep probabilistic or fuzzy categories Initially learn some aspects or ex of different concepts and then classify a new instance on the basis of how similar it is to the previously learned ex Exemplar Models - That receive most attention are the ones that are the most similar to the item to be classified have the greatest influence on categorization Prototype can be seen easier on a transfer test example rather than original learning The prototype should reliably remind the learner of ex from the correct category These models do not assume that people are able to retrieve individual examples one at a time without confusing them

Chapter 10 In HEAD_TO_HEAD COMPETITION , exemplar models have been substantially more successful than prototype models The data favor exemplar model over prototype models WHY? One of the main functions of classification is to allow one to make inferences and predictions on the basic of partial info Relative to prototypes , exemplars tend to be conservative about discarding info that facilitates predictions Exemplar models support predictions and inferences bittern than do prototype models Prototype is average exemplar, therefore loses info about the specific category members that were seen Activating one exemplar is likely to activate others that are similar to it Exemplar models preserve info more than prototype models, info that people seem to be able to use There have been a number of different learning models that assume different sorts of learning processes The legacy of exemplar models is that they have a number of effective characteristics that allow them to account for key phenomena, and future generations of models embody these characteristics Classifying on the basis of the single most similar example is not far from the optimal classification Between-Category Structure o Hierarchical Structure and the Basic Level - Categories are often categorized taxonomically at different levels of abstraction o Intermediate level of abstraction, which seems to provide the label that we would use as a default is called "The Basic Level" o This level appears to be salient and psychologically privileged o The basic level resides at a middle level of abstraction and it appears to be especially salient and psychologically privileged o More abstract are called superordinate's and more similar are called subordinates o Another idea is that levels higher than the basic level have different purposes o Non-Hierarchical Categories Many categories that we use do not fit into a taxonomic category If there is no clear hierarchy then by definition there is no basic level One can think of the various categorization schemes as competing for attention Two factors that influence category access are frequency and recency with which a category has been used Scull and Wryer illustrates the point that category accessibility has an important influence on how social info is encoded and interpreted There is even evidence that activation of one category leads to the inhibition of competing social categories Does Similarity Explain Categorization o Formal models of similarity allow similarity to be quite flexible o Influential Contrast Model = Similarity relationships depend heavily on the particular weights given to individual properties or features o Relative weighting of a feature varies with the stimulus context o Knowledge about these properties such as heart breathing is just these sorts of properties serves to organize children's conceptual and semantic dev

Chapter 10 One could argue that without constraints on what is to account as a feature, any two things may be arbitrarily similar or dissimilar o One possibility that provides a basis for determining categories is that children are not as flexible about similarity as adults o Dev increase in the tendency of children to weight dimensions differentially o Children rapidly learn what types of similarity matter in particular contexts o Structural approaches to similarity such as Geon theory may be more constrained than featural approaches and consequentially may provide more stability o Even if similarity can be more constrained, it still may not explain categorization o Things that look alike superficially often tend to be alike in other deeper ways o Overall similarity is a good but fallible guideline to category membership o Similarity affects categorization and categorization affects similarity as well o There is increasing evidence that the feature or building blocks of categories are not hardwired into the nervous system rather can be modified by exp o Findings suggest that categorization and similarity are closely intertwined, so much so that one could as well say that categorization causes similarity as the converse o SUMMARY - similarity may be the outcome or by-product rather than the cause of categorization Concepts as Organized by Theories o Org of concepts is knowledge-based and driven by intuitive theories about the world o The relation between a concept and an example is analogous to the relation between theory an data o Classification is not based simply on a direct matching of properties of the concept with those in the example but rather requires that the example have the right "explanatory relationship" o the theory org the concept o Classification may be more like an inference process than like a similarity judgment o GOOD of Theory based approach is that it begins to address the question of why we have the categories we have or why categories are sensible o One could readily make judgments about whether new examples belonged to the category, judgments, that would not be based on overall similarity to category members o Children's bio theories guide their conceptual dev o Concepts might be knowledge-based suggests a natural way in which concepts may changenamely, through the addition of new knowledge and theoretical principles Putting Similarity in Its Place o Theories may affect similarity o Similarity may also act as a constraint on theories o The impact of perceptual similarity on the dev of causal explanations is evident in the structure of people's everyday theories o Homeopathy - being that causes and effects tend to be similar o Strong support for : Resemblance is a fundamental conceptual tool of everyday thinking in all cultures, not just so-called primitive cultures o People tend to assume that causes and effects should be of similar magnitude o It is important to recognize that homeopathy often points us in the right direction o Psychological Essentialism - One way of integrating similarity and explanation o Main idea from ^ = People act as if things have essences or underlying natures that make them the thing that they are o Although these characteristics are less reliable than genitive evidence, they are far from arbitrary
o

Chapter 10 Essentialism refers to how people approach the world If people assume the prior line , it may be a good strategy for learning about the world Recall that categorization faces computational complexity problems and that organisms have a strong need to make correct and useful predictions o People adopt an essentialist strategy , the generalization that things that look alike tend to share deeper properties because its usually an effective strategy o Even young children seem to use notions of essence in reasoning about biological kinds o People are sensitive to causal relations and that the notion of an essence isn't needed o Attention has also been directed at different ways of integrating similarity-based and explanation-based learning Do Different Principles Apply for Different Kinds of Concepts o Cognition is organized in terms of distinct domains each characterized by innate constraints or skeletal principles of dev o It is transparently obvious that there are different kinds of categories, at least in the everyday sense of different and kinds o Categories can differ in structure , in the processing principles associated, or in principles that are tied to spec contents o Notion of domain-specificity = The idea is that by looking only at principles that apply to all categories, we may be missing the important principles that apply only to an important subset of categories o Continued research interest has been thrown between nouns and verbs which may differ in structure o Nouns refer to clusters of correlated properties that create bounded chunks of perceptual experience o The distinction between nouns and verbs is more subtle than we have implied o Even more attention ahs been directed at the idea of domain-specificity o The notion of domain-specify has served to organize a great deal of research on conceptual dev o Only later on in life to children organize their knowledge into a less human centric, biological form where human beings are simply one animal among many
o o o

Use of Categories in Reasoning When standard category task is mudded to include inference learning the pattern of results often changes To understand concept learning we need to examine a number of conceptual functions and not just category learning Goals and Ad Hoc Categories o Goal-derived Categories - Organization of categories spontaneously constructed in the service of goals which we will refer to as ad hoc categories o Rather similarity to an ideal o If goal-derived categories are used repeatedly and consistently they may become more stable in memory o That goals and the reasoning processes associated with them can affect category structure Conceptual Combination - Concepts provide the building blocks or mental tokens that enable the construction of new concepts from old ones o ^Conceptual Combination - allows us to produce a virtually unlimited set of new concepts o How are people able to understand these novel concepts o A Complete theory of how people combine concepts does not exist.

Chapter 10 We know that the most straightforward ideas are incomplete In effect, one would be constructing a new prototype and typicality could be judged with respect to this constructed prototype o CC is more complex than the mod model implies o The typicality of combined concepts cannot be predicted form the typicality of two separate concepts o One way for prototype model to handle results is to argue that people have different prototypes for different types of spoons o This strategy also creates problems because one cannot have a separate prototype would lead to many combinations o Another problem is that combined concepts may have properties that do not appear as properties of eater constituent concept o People may complicate things by using a variety of strategies for comprehending combined concepts Categories and Induction o Induction - Use of some knowledge to draw inferences or expectations about other , more general situations o Main idea is that judgments are based on both similarity to examples and knowledge of category membership o Model is formulated in terms of argument strength going from premise to premise to the conclusion o Argument strength is assumed by 2 factors : The similarity of premise categories to the conclusion category and similarity of premise categories to the lowest-level category o Similarity , typicality, and diversity effects can be mea to disappear or even reverse using examples where undergraduates do have relevant knowledge o Psychologists have started exploring the use of categories in reasoning
o o

Summary Concepts serve to organize our mental life